A journal's impact factor is roughly the mean number of citations its recent articles received in a given year.
A journal's impact factor is roughly the mean number of citations its recent articles received in a given year. It is a property of the journal across all its output, and it is calculated from a distribution that is extremely skewed.
Within any journal a small minority of papers collect most of the citations. The average is therefore unrepresentative of a typical article, and inferring anything about a specific paper from its venue's average is a straightforward statistical error.
Citation rates vary enormously between fields for reasons unrelated to quality: how fast a field moves, how many researchers it has, how many references a typical paper carries. Comparing across fields compares cultures rather than standards.
A paper can be cited because it is wrong, because it is a convenient example, or because it is the standard reference for a method. Citation counts measure attention, and attention and correctness are only loosely related.
Review articles attract citations, so a journal publishing more of them raises its figure without changing the quality of its research articles. Other editorial choices have similar effects, which is why the metric is contested within publishing itself.
Comparing journals within one field, as one input among several, when choosing where to look. That is close to the original purpose and some distance from how it is now used.
Citations to the specific paper, and attempts at replication, say more about one piece of work than its venue's average ever can. Both are available and neither is as convenient.
Deciding a finding is reliable because of where it appeared skips the only question that matters, which is what the study actually did. A strong study in an obscure journal is better evidence than a weak one in a famous journal.
Within any journal a small minority of papers attract most citations, so the mean describes almost no individual article. Inferring a specific paper's quality from its venue's average is a straightforward statistical error.
Citations to the specific paper, and whether anyone has replicated it, say far more about one piece of work than its journal's average ever can. Both are available and neither is as convenient.
The claims on this page about a journal-level average, routinely used to judge individual papers it says nothing about can be checked against sources outside this site: